Yusiko/qwen3.5-prompter

🧠 Qwen3.5-4B Prompter — GGUF

A multilingual prompt engineer model fine-tuned on Yusiko/prompter — a 5,000-sample dataset covering 10 languages and 7 domains.

Given any short, vague user input, this model expands it into a fully structured, production-ready prompt with role assignment, context, step-by-step instructions, output format, and quality standards — following Google's Prompt Engineering Whitepaper best practices.

🚀 Trained 2x faster with Unsloth · Exported to GGUF · Ready for Ollama & llama.cpp

---

📦 Available Files

File	Quantization	Size	Use case
Qwen3.5-4B.Q4_0.gguf	Q4_0	~2.54 GB	💡 Recommended — fast, efficient
Qwen3.5-4B.BF16-mmproj.gguf	BF16	larger	🔬 Higher precision

---

🚀 Quick Start

Ollama

ollama run hf.co/Yusiko/qwen3.5-prompter

llama.cpp

# Text-only
llama-cli -hf Yusiko/qwen3.5-prompter --jinja

# Multimodal
llama-mtmd-cli -hf Yusiko/qwen3.5-prompter --jinja

Python (llama-cpp-python)

from llama_cpp import Llama

llm = Llama.from_pretrained(
    repo_id="Yusiko/qwen3.5-prompter",
    filename="Qwen3.5-4B.Q4_0.gguf",
    n_ctx=2048,
)

response = llm.create_chat_completion(
    messages=[
        {
            "role": "user",
            "content": (
                "Below is an instruction that describes a task, paired with an input "
                "that provides further context. Write a response that appropriately "
                "completes the request.\n\n"
                "### Instruction:\n"
                "As a prompt engineer, transform this simple input into a fully detailed, professional prompt\n\n"
                "### Input:\n"
                "Write a Python function\n\n"
                "### Response:"
            )
        }
    ],
    max_tokens=512,
    temperature=0.7,
)
print(response["choices"][0]["message"]["content"])

---

💡 Prompt Format (Alpaca)

Below is an instruction that describes a task, paired with an input that provides further context.
Write a response that appropriately completes the request.

### Instruction:
As a prompt engineer, transform this simple input into a fully detailed, professional prompt

### Input:
{your simple prompt here}

### Response:

⚠️ Always use string concatenation — not .format() — when building prompts programmatically. The model's outputs contain {curly braces} that will cause KeyError with Python's string formatter.

---

🎯 What This Model Does

Input: a short, vague prompt

Write a Python function

Output: a complete, structured, professional prompt

## System Prompt
You are a senior software engineer with 10+ years of Python experience.
Your task is to write a Python function with production-quality standards.

## Role & Context
Act as a senior engineer conducting a thorough implementation session...

## Step-by-Step Instructions
1. Clarify requirements, edge cases, and constraints before writing any code
2. Design the interface and data structures first, then implement logic
3. Write the implementation with comprehensive inline documentation
4. Add input validation and robust error handling for all edge cases
5. Write unit tests covering happy path, edge cases, and error scenarios

## Output Requirements
- Implementation: Complete, working code with no placeholders
- Documentation: Inline comments explaining non-obvious logic
- Tests: At minimum 3 test cases (happy path, edge case, error case)
...

---

📊 Training Details

Field	Value
🤖 Base model	Qwen/Qwen3.5-4B
🗂️ Dataset	Yusiko/prompter
📦 Dataset size	5,000 samples
🌍 Languages	10 (az, en, tr, ru, de, fr, zh, ar, es, ja)
🎯 Method	QLoRA (rank=16, alpha=16)
⚙️ Framework	Unsloth + TRL SFTTrainer
💻 Hardware	NVIDIA RTX5070 12GB
🧮 Optimizer	AdamW (PyTorch)
📐 Seq length	1024 tokens
🔢 Batch size	1 × 8 grad accum = 8 effective
📉 LR scheduler	Cosine
🔁 Training steps	500
🏷️ Export format	GGUF Q4_0

---

🏗️ Dataset Overview

The Yusiko/prompter dataset contains 4 output types, each following Google's Prompt Engineering Whitepaper:

Type	Count	Description
🔷 Standard	~3,280	Role + system + contextual prompting
🔶 Few-shot	~1,000	2 examples shown before the main task
🔹 Chain-of-Thought	~460	Step-by-step reasoning structure
🔸 Step-back	~260	General principles → specific implementation

Domains covered: Coding · Writing · Analysis · ML/AI · DevOps · Data Engineering · Business Strategy

---

⚙️ Hardware Requirements

Setup	VRAM / RAM	Speed
GPU (Q4_0)	4–6 GB VRAM	Fast
CPU only (Q4_0)	~6 GB RAM	Moderate
Apple Silicon (Q4_0)	~6 GB unified RAM	Fast via Metal

---

📜 Citation

@model{yusiko_qwen35_prompter_2025,
  author    = {Yusif},
  title     = {Qwen3.5-4B Prompter: Multilingual Prompt Engineering Model},
  year      = {2025},
  publisher = {HuggingFace},
  url       = {https://huggingface.co/Yusiko/qwen3.5-prompter},
  dataset   = {https://huggingface.co/datasets/Yusiko/prompter}
}

---

🙏 Acknowledgements

• Unsloth — 2x faster fine-tuning, GGUF export
• Google Prompt Engineering Whitepaper — Lee Boonstra et al., Feb 2025
• TRL — SFTTrainer + SFTConfig
• Qwen Team — Qwen3.5 base model

---

Built with ❤️ by Yusif · Apache 2.0 License